Extended Abstract. In the past decade various studies were done on the usage and influence of the internet, technology, and social media in society [1, 2, 3]. These studies indicate that social media has become a popular means of communication and therefore forms a significant part of what determines the views and opinions of society [4]. It is therefore increasingly important to analyse and understand social media networks and information flows. Understanding the ways that online communities form and communicate, as well as the different underlying network structures, could possibly assist with identifying key influencers as well as the formation and dynamics of social networks and communities [5]. The information can be utilised, for example, in the design of necessary interventions, or the identification of targeted campaigns [6]. The aim of this study was to extend previous research [7] and investigate the potential network structures that are formed within the South African Twitter community around the 2019 elections by means of knowledge graphs with an existing toolset namely NodeXL. Previous research and findings suggest that social network maps of Twitter communities can be analysed and visualised to provide insight into the setting of social media. The network maps point out previously unknown information about the persons and topics that drive conversations and group behaviour [8]. The main research question answered by this study can be stated as follows: Are there any distinct network patterns that can be detected within the South African Twitter community, specifically concerning recent political concerns? Twitter datasets on the SA elections during three specific periods were collected, namely, the last week that political campaigning was allowed (28 April 2019 - 4 May 2019), Election day on 8 May 2019, and Inauguration day on 24 May 2019. The datasets were imported into NodeXL, and unique entities, or vertices, were determined. The vertices were grouped into clusters using the Clauset-Newman-Moore algorithm that analyses how the vertices are connected to one another. A range of measures of the graph was calculated for the overall network, as well as for each vertex in the network. Each of the network metrics summarizes a different aspect of the size and shape of the overall network and the location and connection properties of each entity in the network graph. For all three datasets groups with less than ten vertex connections as well as neighbourless edges were removed from the graph in order to focus the network
analysis on the core network. Subgraph images were created for each vertex before the overall network was visualized using the NodeXL visualisation feature. Figure 1 below depicts the graph of Dataset 3 – Inauguration Day (23-24 May 2019).
All three of the datasets that were analysed depicted community clusters that have a large number of disconnected entities that contribute to the network by mentioning the topic but does not link to other entities within the network. Both the main, larger groups as well as the smaller, sub-groups are highly interconnected and only share a small percentage of connections across groups. Community clusters often reveal diversity of opinion on a specific topic that exists within a society. Since South Africa has multiple political parties, multiple groups are formed, both on the general subject of the elections, as well as centred around specific political groups or influencers. In Datasets 1&2 groups shared a common topic but had a different focus, while the groups in Dataset 3 remain segregated.
In conclusion Twitter has become a popular means of communication and
conversations create networks with identifiable groups and connections as users reply to and
mention one another. Mapping social networks can assist in understanding the different
ways that individuals form communities and organize online. Previous research
identified six distinct patterns in the social structures depending on the topic in question [